Merge pull request #282 from vidartf/optimize

Add caching for mimedata comparison

nbdime/diffing/notebooks.py

@@ -3,21 +3,25 @@
 # Copyright (c) Jupyter Development Team.
 # Distributed under the terms of the Modified BSD License.
 
-from __future__ import unicode_literals
-
 """Tools for diffing notebooks.
 
 All diff tools here currently assumes the notebooks have already been
 converted to the same format version, currently v4 at time of writing.
 Up- and down-conversion is handled by nbformat.
 """
 
+from __future__ import unicode_literals
+
 import operator
 import re
 import copy
 from collections import defaultdict
 from six import string_types
 from six.moves import zip
+try:
+    from functools import lru_cache
+except ImportError:
+    from backports.functools_lru_cache import lru_cache
 
 from ..diff_format import MappingDiffBuilder, DiffOp
 
@@ -49,6 +53,7 @@
 #       an argument instead of separate functions.
 
 
+@lru_cache(maxsize=1024, typed=False)
 def compare_text_approximate(x, y):
     # Fast cutoff when one is empty
     if bool(x) != bool(y):
@@ -90,6 +95,15 @@ def compare_base64_strict(x, y):
     return x == y
 
 
+@lru_cache(maxsize=128, typed=False)
+def _compare_mimedata_strings(x, y, comp_text, comp_base64):
+    # Most likely base64 encoded data
+    if _base64.match(x):
+        return comp_base64(x, y)
+    else:
+        return comp_text(x, y)
+
+
 def _compare_mimedata(mimetype, x, y, comp_text, comp_base64):
     mimetype = mimetype.lower()
 
@@ -108,12 +122,7 @@ def _compare_mimedata(mimetype, x, y, comp_text, comp_base64):
     # TODO: Compare binary images?
     #if mimetype.startswith("image/"):
     if isinstance(x, string_types) and isinstance(y, string_types):
-        # Most likely base64 encoded data
-        if _base64.match(x):
-            return comp_base64(x, y)
-        else:
-            return comp_text(x, y)
-
+        _compare_mimedata_strings(x, y, comp_text, comp_base64)
     # Fallback to exactly equal
     return x == y
 
@@ -357,17 +366,20 @@ def diff_single_outputs(a, b, path="/cells/*/outputs/*",
     if a.output_type in ("execute_result", "display_data"):
         di = MappingDiffBuilder()
 
+        # Separate data from output during diffing:
         tmp_data = a.pop('data')
-        a_conj = copy.deepcopy(a)
-        a.data = tmp_data
+        a_conj = copy.deepcopy(a)  # Output without data
+        a.data = tmp_data          # Restore output
         tmp_data = b.pop('data')
         b_conj = copy.deepcopy(b)
         b.data = tmp_data
+        # Only diff outputs without data:
         dd_conj = diff(a_conj, b_conj)
         if dd_conj:
             for e in dd_conj:
                 di.append(e)
 
+        # Only diff data:
         dd = diff_mime_bundle(a.data, b.data, path=path+"/data")
         if dd:
             di.patch("data", dd)

nbdime/profiling.py

@@ -40,6 +40,7 @@
 import time
 import contextlib
 from tabulate import tabulate
+from functools import wraps
 
 
 def _sort_time(value):
@@ -55,18 +56,31 @@ def __init__(self, verbose=False, enabled=True):
 
     @contextlib.contextmanager
     def time(self, key):
+        if not self.enabled:
+            yield
+            return
         start = time.time()
         yield
         end = time.time()
         secs = end - start
-        if not self.enabled:
-            return
         if key in self.map:
             self.map[key]['time'] += secs
             self.map[key]['calls'] += 1
         else:
             self.map[key] = dict(time=secs, calls=1)
 
+    def profile(self, key=None):
+        def decorator(function):
+            nonlocal key
+            if key is None:
+                key = function.__name__ or 'unknown'
+            @wraps(function)
+            def inner(*args, **kwargs):
+                with self.time(key):
+                    return function(*args, **kwargs)
+            return inner
+        return decorator
+
     @contextlib.contextmanager
     def enable(self):
         old = self.enabled
@@ -88,8 +102,8 @@ def __str__(self):
         for key, data in items:
             time = data['time']
             calls = data['calls']
-            lines.append((key, calls, time))
-        return tabulate(lines, headers=['Key', 'Calls', 'Time'])
+            lines.append((key, calls, time, time / calls))
+        return tabulate(lines, headers=['Key', 'Calls', 'Time', 'Time/Call'])
 
 
 timer = TimePaths(enabled=False)
@@ -98,7 +112,8 @@ def __str__(self):
 def profile_diff_paths(args=None):
     import nbdime.nbdiffapp
     import nbdime.profiling
-    nbdime.nbdiffapp.main(args)
+    with nbdime.profiling.timer.enable():
+        nbdime.nbdiffapp.main(args)
     data = str(nbdime.profiling.timer)
     print(data)
 

setup.py

@@ -258,6 +258,7 @@ def run(self):
 
     ':python_version == "2.7"': [
         'backports.shutil_which',
+        'backports.functools_lru_cache',
     ],
 }